Skip to main content

Full text of "Industry Report. The Gray Sheet. March 3 1972. Vol. 7, No 9"

See other formats

------------------------------<page break>-----------------------------
During the early t96Os, various research projects -- most of them funded by Arpa 
(the Advanced Research Projects Agency of the Defense Department) -- brought the 
phrase "timesharing" to the lips of computer industry buffs. This work paved the 
way for a business that will hit $400 million this year, up 33% over 1971. 
Now, Arpa is quietly touting another concept -- resource sharing -- that may 
lead to vast extensions of the early timesharing concept. By tying together a 
large number of computers through communications "nodes" -- themselves little 
computers -- the Arpa researchers have come up with a new kind of network . 
a community of computers that share vast calculation power, massive data stores, 
and specialized software. 
The system, called Arpanet, already contains over 25 nodes (chart, page 3) 
and is looking for more users. Its final impact on the computer industry is un- 
measurable at this time, but could be significant; the ťoncepts behind Arpanet 
reflect many ideas in the mainstream of advanced planning for distributed com- 
puting and data transmission. Similar systems could be built for immediate com- 
merciat use. 
In a nutshell, Arpanet links together computer resources so that they are 
used at maximum efficiency. Three aspects are involved: 
▀ Load sharing brings a wealth of computer power to Arpa researchers 
on demand. Weather Bureau scientists can use the Sigma 7 at UCLA; 
MIT students, if they overload the school's PDP-10 during peak morn- 
ing hours, can access a similar machine at Utah for the overflow, 
with the reverse possible after lunch to smooth the system load. 
▀ Data base sharing prevents duplication of efforts. All researchers 
have access to the massive data bases generated at other sites, and 
this capability will increase next year when Itliac IV and its tril- 
lion-bit store join the network. 
▀ Software sharing cuts one of the most significant costs in using a 
computer, and will provide transfer of human skills. Arpa researchers 
can access any specialized software on any system rather than develop 
their own or adapt it to suit a particular type of hardware. 
EDP Industry Report and Market Review is published twenty-four times each year by the International Data Corporation, 60 Austin 
Street, Newtonville, Mass. 02160. Entire contents Ş 1972 by the international Data Corporation. James Peacock, Managing Editor; 
Fred Anderson, Michael Merritt, Nancy Scull, Associate Editors; C, Oakley Mertz, Research Analyst; Arthur Whetan, Circulation 
Manager, Patrick J. McGovern, President, International Data Corporation; William Leitch, Vice President/Subscription Services; Walter 
Misdom, Vice President/Industry Analysis. Editorial ideas and information are collected from numerous industry experts, and are 
researched and formulated by members of the editorial staff. The information and opinions contained in each issue are based on the 
best information available, but their accuracy and completeness cannot be guaranteed. Subscriptions: One Year, $75. Single copies, 
priced as indicated on inside pages, air-mailed same day requested, Prices for multiple copies available on request. Telephone (617) 
969-4020, Reproduction without written permission is completely forbidden. 
------------------------------<page break>-----------------------------
JFOR ]972 
With such an interconnected system, virtually all the independent research 
efforts to advance computing in the U.S. are brought together under one roof, as 
it were, and will have a common orientation. No duplication of effort will be 
required. Yet all researchers interested will be able to access a one-of-a-kind 
machine such as Illiac IV, which joins the network next March, or work together 
on a common language for the system. This fusion of human efforts combined with 
almost human communication between machines is what Arpa means by resource sharing. 
How Arpanet works 
As the map above shows, Arpanet is a closed circuit network of 50 kilobit 
communications lines that connect over 25 research centers across the nation. 
Interconnection by conventional means would have required separate lines between 
each pair of nodes -- in a "star" fashion -- and the failure of any single line 
would have isolated those two nodes from each other. With just 40,000 miles of 
data transmission pipeline, however, Arpanet virtually guarantees communications 
capability between any two nodes at all times. If one link is unusable, another 
route always exists; it is located automatically by an element of the system. 
The heart of the network is the node, and the network has many hearts. This 
little computer called an IMP (Interface Message Processor) receives and trans- 
mits messages according to a standard set of rules called a protocol. The IMP 
results from the system store-and-forward technology developed by Bolt Beranek and 
Newman Inc., also one of the timesharing originators. The node hardware is based 
on a Honeywell DP-516 or DDP-316 minicomputer that is modified, installed, and 
maintained by BBN. Each of the IMPs, priced at $60,000, contains the system rules, 
or protocol, and they reload each other in the event of failure or removal from 
the communication network for maintenance. 
March 3, 1972 -2- EDP INDUSTRY REPORT 
Rrucon rbn. Addition copies airmed same day requested. Price: Subscribers $3 per copy; others $5 C (617) 969-4020 
------------------------------<page break>-----------------------------
The I is what actually connects the computers (Hosts) to the network. If 
a researcher at Carnegie, for example, wants to access a data base at Rand, his 
Host PDP-t0 must transmit an appropriate message to its IMP. The IMP, in turn, 
acknowledges receipt of the message, translates it into a standard code, selects 
the fastest route -- considering the distance involved, the status of the lines, 
and load factors along the way -- and transmits the message to the next IMP in 
the route. 
As the message proceeds to its destination, each intermediate IMP tells the 
previous one the message was received. Then, it checks the selected route to make 
sure it is still the best way to the final destination, and continues the trans- 
mission process. Ultimately the message arrives at its destination. In this ex- 
ample, for instance, the final IHP decodes the message and transmits it to the 
Rand Host. That Host then starts a confirmation message on its way back to the 
Carnegie IIost to acknowledge receipt. 
Amazingly, this entire process takes no more than 0.2 seconds. In fact, when 
the system was demonstrated to EDP Industry Report at the BBN control center re- 
cently, we saw no noticeable difference in transmission time from one Host to a- 
other. It didn't seem to matter whether we interacted with a Host in Utah or the 
Project MAC computer center down the street at MIT in Cambridge. The only appar- 
ent delays stemmed from complex operating systems in the Host computers themselves. 
The coding and decoding tasks performed by the IMPs are no small feat. In 
fact, it is precisely for this reason that so many different types of computers 
can participate as Hosts. The schematic above gives some indication of the enor- 
mous complexity of the system; each different Host -- from Burroughs, DEC, IBM, 
>[DS -- communicates in a different code. The IMP simply doesn't care. 
March 3, 1972 -3- EDP INDUSTRY REPORT 
Repruction rbidden. Addition copies airmailed same day requested. Price: Subscribers $3 per copy; others $5 ('all (617) 969-4020 
------------------------------<page break>-----------------------------
Bring on the terminals 
A more recent development within Arpanet extends the resource sharing con- 
cept still further and opens the available resources to multitudes of new users. 
Working as development coordinator for the system, Bolt Beranek and Ne,an has 
added to the IMP sufficient capability for interfacing a terminal directly to 
Arpanet. This modified IMP is called a TIP (for Terminal Interface Processor) 
and includes multiplexing capabilities for up to 63 terminals, costs $120,000. 
Now, for example, weather researchens at ETAC (Environmental Technical Ap- 
plications Center) in Washington, D.C. can use computer power, software, and data 
at the University of Utah rather than having to duplicate these resources in Wash- 
ington. Also, as the logical map (page 3) shows, Global Weather Central will soon 
be hooked into the system through another TIP. 
The network, today, contains a relatively complete array of computer power, 
but is operating at only a few percent of capacity. Last fall, for example, the 
675,000 message-per-day traffic was just over 2% of the 30 million capacity. So 
Arpa sees the next rash of new nodes as being TIPs rather than additional IMPs 
and associated Hosts. 
How much does it cost? 
The greatest single advantage in tying together computers in this fashion 
-- whether for research as in Arpanet or for certain business-related uses that 
are projected -- is the tremendous economy. Extensive studies undertaken for 
Arpa show that the capacity and cost of such distributed networks are "remarkably 
insensitive" to the distribution and destination of traffic. The total traffic 
is the only important arameter. Thus, says Arpa, "it is appropriate to charge 
for traffic initiated at a node, based on the cost of increasing the total capa- 
city of the network by that amount." Charges, thus, could be much like those of 
the Postal Service, 'with message length (weight) more important than distance. 
At present, an equal share of the communication line cost is being allocated 
to each node, so the cost for Arpanet users is about $4,800 per node per month -- 
$3,100 for communications and $1,700 for leasing a minimal IMP. Use of a TIP in- 
creases the cost by $1,600 a month. 
But this will change. As soon as feasible, Arpa plans to charge for traffic 
initiated at a node. The actual charges will depend on the loading that can be 
expected for the system, since it will naturally not operate at peak capacity day 
in and day out. Arpa's target rate, based on 36% loading, is an estimated 30C 
per megabit in addition to equipment rental. 
A few problems 
As can be expected for any new development in computing, a number of problems 
-- legal, technical, and human -- will prevent this concept from mushrooming full 
bloom into a series of networks connecting most of the computers in the country. 
Five or ten years down the pike, however, significant changes in computer system 
organization are expected. And in the light of such a timetable, the anticipated 
problems don't seem insurmountable. In fact, they are quite similar to those al- 
ready being faced by a number of conputer users and, because of the resource shar- 
ing concept, if they are solved once, they are solved for all. Some of today's 
"challenges" are: 
March 3, 1972 -4- EDP INDUSTRY REPORT 
Rruction rbidden. Addition copies mlcd sne day requested. Price: Subscribers $3 per copy; others $5 Call (617) 969-4020 
------------------------------<page break>-----------------------------
▀ Computer/communications interrelationships have been studied exhaust- 
ively by the Federal Communications Commission (EDP/IR Apr.9,19?0), 
but all the questions weren't answered. It would appear the Arpanet 
would be what the FCC calls a "hybrid" system -- "an offering of ser- 
vice which combines Remote Access data processing and message-switch- 
ing to form a single integrated service." Such systems, the [CC says, 
need be regulated if and only if message switching is the main purpose 
of the system and data processing is an incidental and integral part of 
the system. It looks to EDP/IR as if additional definition -- or at 
least a firmer ruling -- will be in he cards. 
Host-Host protocol is an on-going research project within the Arpanet 
members to allow graphics, block transfer of data, and file transfers. 
A colamon macro language doesn't seem to be necessary at this point in 
time, but it would appear necessary before similar systems gain wide- 
spread commercial usage. 
▀ The privacy question will likely raise its head high as vast networks 
are established. The Arpanet really doesn't affect the invasion of 
privacy issue per se, but it could be a strong factor in bringing the 
question to the surface faster and forcing answers sooner than would 
be expected under current conditions. 
▀ On the other hand, the sharing or resale of common carrier facilities 
cannot be done for profit. Thus, if several service organizations such 
as Control Data, GE, and Tymshare set up a jointly owned network (see 
below), provision for this current FCC requirement might be a problem. 
The potential impact 
As suggested earlier, many commercial "communities" today have characteris- 
tics and needs for which an Arpanet-like linkage might be useful today. There's 
no question but that some of these will evolve: 
The timesharing industry itself could gain two primary advantages from set- 
ting up a jointly-owned system: 
+ The network could provide a highly reliable communications system 
for interconnecting remote locations and central computers; 
+ By mutual agreement, services within the network could draw on each 
other's powers and resources for backup and a more complete spectrum 
of service capabilities. 
The large service companies mentioned above have their own private communications 
networks, but smaller companies cannot afford the luxury of such mammoth develop- 
ment efforts or the full time broadband communications links required. Also, 
even for these large companies, individual networks do not operate at nearly the 
efficiency that Arpanet sees right now. 
Inter-airline reservations would be greatly simplified by interconnection 
through an Arpanet cousin. If one airline could not supply the requested seat, 
it could automatically query the network to find out who could do it, then write 
the ticket and perform the appropriate accounting transactions all without 
any further human intervention. 
March 3, 1972 -5- EDP INDUSTRY REPORT 
Reproduction forbidden. Additional copies ai:rnailed same day requested. Price: Subscribers $3 per copy; others $5 ('all (617) 969-4020 
------------------------------<page break>-----------------------------
Hospital patient records become increasingly dispersed as American society 
becomes more mobile. The Arpanet concept, however, provides a means for any par- 
ticipating hospital to query all other hospitals until a given patient's records 
are found. The local Host could then update the records on the remote Host rather 
than duplicating storage. 
Inter-library tendin systems could extend the resources of the nation's 
numerous libraries far beyond their present limits. Many companies have small, 
specialized libraries and draw on others for information they do not maintain 
themselves. At present, however, a librarian in one of these organizations must 
search manually for a library that has the publication required. A computerized 
network of library catalogs could track down the appropriate material far more 
quickly and efficiently. 
Federal and state governments have numerous applications for which such a 
network could cut computer investment and line charges significantly. Some ini- 
tial efforts at cooperation -- such as between motor vehicle bureaus -- are un- 
derway, but many other areas haven't even been considered. 
Looking more than a few years into the future, EDP/IR can forsee even the 
possible specialization of computers by function. Certain computers, for ex- 
ample, might perform only payroll calculations for all network participants; 
another network might keep track of corporate shareholder records; and a third 
might automat{cally compute everyone's income tax and bill the individual for 
the amount due. Many of these would work with banks in the transfer of money. 
Arpa has expressed similar possibilities for the selection of computers: 
Soon after a network with a dozen or more reliable comouter services be- 
comes available, many institutions will find it far more economic to ob- 
tain their computing services from a selected set of these remote systems, 
rather than run their own computer center. 
For example, take the case of an institution about ready to upgrade its 
facility. One choice would be zo obtain a medium scale, general purpose 
batch system, admittedly a compromise for their large numerical users and 
timesharing users, but the best single system that they can afford. Al- 
ternatively, they could buy no new machine and obtain access to several 
of the systems on the network through a TIP. This approach permits their 
large numerical users to use a large "number-cruncher," their statistical 
and payroll users to access a large scale general purpose system, and 
their interactive users to have Teletype or graphic console access to a 
good timesharing system. Overal2 the cost of each service is less than 
it would have been on a dedicated G-P computer by factors of between two 
and ten. Also, they can buy just the capacity they need and expand smoothly 
rather than having to pay for an oversize machine for a year or two. 
That's hardware sharing, and the same type of advantages are expected from 
data base sharing and software sharing. Once these are going, says Arpa, there 
will develop. other important network applications, "all of which require a large 
viable network before they become important in their own right. These include 
teleconferencing, publishing, library services, and office paperwork filing and 
distribution. Ten to twenty years from now these applications may well dominate 
computer usage and network usage but they are not likely to be important factors 
for at least five years." EDP/IR, along with Arpa, is convinced this is the wave 
of the future. And, in their petty pace, things are headed in this direction. 
March 3, 1972 -6- EDP INDUSTRY REPORT 
Rrucon forbn. Addition copies airmed same day requested. Price: Subscribers $3 per copy; others $5 CI (617) 969-4020 
------------------------------<page break>-----------------------------
With all the talk of "saturation" that goes around from time to time, EDP/IR 
asked the International Data Corporation, its publisher, to compile some relevant 
data. The figures below for general-purpose (Group A) computers on___ Z are based 
on the January, 1972 U.S. Computer Installation Data Pile maintained by IDC. Fig- 
ures for tax reporting units -- plant sites for industrial companies -- are ex- 
trapolations based on 1968 data from 1970 Statistical Abstracts, U.S. Department 
of Commerce. This is the latest data available to IDC at presstime. 
No. Employees Estimated Estimated 
per tax No. of total No. 
reporting reporting of Group A 
unit units computers 
t- 19 
20- 49 
50- 99 
t00- 249 
250- 499 
500- 999 
Over 5000 
3,260 000 
95 400 
54 50O 
3 620 
% with 
t or more 
No. of sites 
with one or 
1 630 
3 450 
4 500 
8 290 
8 000 
5 330 
3 260 
1 630 
3 620 
4 950 
9 810 
7 460 
6 520 
2 745 
SUBTOTAL 3,713,537 0.96 35,795 49,050 
Federal Government 
State & Local Government 
1,325 3,585 
27310 2725 
TOTALS 39,430 55,360 
The figures above indicate that, on average, there are 1.4 general-purpose 
computers per site in the U.S. That average is slightly misleading, however, be- 
cause 75% to 80% of all the sites have only one computer. The remainder, as a 
result, have almost 3 computers per site. 
And this computer concentration is even more pronounced when the fact that 
some companies have many sites is considered. Because of this concentration some 
40% of the computers in the U.S. are operated within only 700 or so organizations. 
Thus, as would be expected, computers follow wealth and high employment. The fig- 
ures above indicate this concentration: Even though just under 1% of all tax re- 
porting units have a computer, the penetration jumps to about 8% when the first 
category -- one-to-twenty employees -- is eliminated from the averages. 
There are many opportunities for installing small computers at the number of 
small tax reporting units without EDP. But the way in which vast numbers of these 
organizations will receive computer power is probably going to be via terminals 
connected to a network of some sort. This is one more reason why commercial op- 
erations such as the one now up-and-running for Arpa researchers (page 1) will be 
so important. And other factors point in this direction. Shipments to new users 
accounted for about one-third of all computer deliveries during the mid-1960s; but 
these peaked in 1968 and 1969 -- at $1.2 billion -- and have started to decline in 
value as well as percentage. By the mid-1970s, shipments to new users in the U.S. 
will be less than 10% of the total, according to IDC projections. 
March 3, 1972 -7- EDP INDUSTRY REPORT. 
RrucHonforbidden. Additioncopicsmedsamcdayxequested. Pice: Subscribexs$3percopy;other$SC(617)969-4020 
------------------------------<page break>-----------------------------
In a move that further suggests its commitment to communications, IBM this 
week announced a controller "with its own processor for improved remote comput- 
ing." The device -- called the 3705 -- is available for shipment this summer, 
and it is being offered "under terms of a new IBM extended term plan." Thus, in 
a single announcement treated routinely, IBM has eliminated a weak link in its 
product line and started people talking again about the possibility of IBM's of- 
fering term leases for additional products -- including mainframes. 
All IBM will say about the new extended-term lease plan is that "we expect 
to announce the ETP to other products, but we will make that decision as the pro- 
ducts come along. At present, we have no plans to add announced products to long 
term leasing plans." 
The plan provides for discounts of about 13% off the regular month-to-month 
rental contract for the product. It provides for an initial 24-month contract 
and gives the user the option of an unlimited number of one-year extensions of 
the contract. And since communications systems are normally installed at con- 
siderable expense and planning, the two-year minimum would seem academic. 
The new product will meet "growing requirements for more flexible and reli- 
able teleprocessing operations." A "combination of logic circuitry and core mem- 
ory," the 3705 is not dubbed a computer by IBM. But it will fill a gap that many 
IBM users, with the airlines at the van, have gone to minicomputer manufacturers 
to supply. IBM says it will take over the task of controlling the access of data 
to and from terminals in a network, thereby relieving the central computer of rou- 
tine tasks such as communication line control, character checking, and buffering 
and line polling. 
The 3705 can handle up to 352 communications lines, twice as many as the 2703 
it replaces. It can handle line speeds ranging from 45.5 bps to 50,000 bps, and 
can support both asynchronous or binary synchronous transmission. The basic unit 
consists of a processor with 16K bytes of core memory. This can be expanded in 
32K increments up to 240K bytes, or characters. The ETP charges range from $1,200 
to $9,500 a month, compared to monthly charges of $1,390 to $10,940 under the 
shorter-term plan. Purchase is $57,000 to $449,000. 
There is nothing very new, technically, about the 3705. Some observers have 
suggested that it is another home for the memory and logic circuits from 360s be- 
ing returned by 370 customers. In any event, it broadens IBM's communications 
offerings and could cut into business previously received by Honeywell or Inter- 
data from customers with 360-based communications systems. 
The 1971 annual report just published by American Telephone & Telegraph Com- 
pany contains a good deal of talk, and some figures, about the data communications 
business as seen by the giant. "Data transmission -- on private line channels and 
on our general message network -- continues, as it has for some years, to be our 
fastest growing service. It brought in revenues of some $650 million in 1971 -- 
and market studies project a ten-fold increase in the decade ahead." AT&T's rev- 
enues from "data services" grew 18% last year, according to a table in the report, 
versus the 21% and 25% figures posted in the prior two years. The compound rate 
for the last ten years has been 26% annually. 
March 3, 1972 -8- EDP INDUSTRY REPORT 
Repruconrbidden. Additioncopiesledsamedayequested. Price: Subscibers$3pecopy:othc$5Ct(617)969-4020 
------------------------------<page break>-----------------------------